[SUPPORT] about hudi

[cbomgit]

I saved a thread dump but here is a screenshot of the thread dump for the executor, which suggests that it's getting hung up here: https://github.com/apache/hudi/blob/release-0.11.1/hudi-client/hudi-client-common/src/main/java/org/apache/hudi/table/action/commit/HoodieMergeHelper.java#L147

[danny0405]

My speculation is that it is because of the false positive of bloom_filter index, while because you are still using 0.11 release, maybe you can try simple index instead.

[cbomgit]

Thanks for the response! Is there any way to confirm the fp rate?

[danny0405]

There is a stage named like "buildPartitionProfile" which would take long time if the fp happens.

[cbomgit]

Thanks. Switching to SIMPLE index doesn't seem to help. Job fails at the following job and stage:

The task logs show OOM error:

2024-08-25T05:22:43.611+0000: [Full GC (Allocation Failure)  36580M->33919M(40960M), 46.7644028 secs]
   [Eden: 0.0B(1792.0M)->0.0B(2048.0M) Survivors: 256.0M->0.0B Heap: 36580.4M(40960.0M)->33919.2M(40960.0M)], [Metaspace: 117312K->117312K(135168K)]
 [Times: user=77.83 sys=0.20, real=46.76 secs] 
2024-08-25T05:23:30.376+0000: [Full GC (Allocation Failure)  33919M->33904M(40960M), 43.9530911 secs]
   [Eden: 0.0B(2048.0M)->0.0B(2048.0M) Survivors: 0.0B->0.0B Heap: 33919.2M(40960.0M)->33904.1M(40960.0M)], [Metaspace: 117312K->113897K(135168K)]
 [Times: user=74.71 sys=0.03, real=43.96 secs] 
#
# java.lang.OutOfMemoryError: Java heap space
# -XX:OnOutOfMemoryError="kill -9 %p
kill -9 %p"
#   Executing /bin/sh -c "kill -9 829
kill -9 829"...

For settings, I set simple index and upsert parallelism to 1500, but it doesn't seem to be using that setting here. Are there any HUDI config knobs you recommend tweaking to increase the throughput here?